Who's afraid of AI photos?

What's scarier: new technology — or liars?

An AI-generated picture of Pope Francis wearing a high-fashion puffy coat went viral last week, and the fact that many were evidently “fooled” by it has been held up as the latest proof point in the growing moral panic over computer-generated images — a picture “highlighting the power and peril of AI,” as CBS put it. 

Personally, I’m a bit skeptical that we’re dealing with anything particularly new here. When I saw the Pope picture, which struck me as obviously fake by virtue of premise more than quality, the first thing that popped into my mind was this grotesque image of Barack and Michelle Obama:

This half-assed Photoshop job will no doubt strike you as ludicrously, even insultingly fake, and yet it apparently “fooled” enough people to earn a full write-up on Snopes.com (“No, this isn't a photograph of Barack and Michelle Obama at a White House function, a wedding, a Halloween party, a St. Patrick's Day party, or any other type of formal affair or party,” they assure). Snopes houses a veritable cornucopia of BS “photos” of Obama that went viral during his presidency, including pictures of Barack wearing an Aleister Crowley t-shirt, Barack open-mouth kissing the prime minister of Britain, and (of course) supposed “pre-op” photos of Michelle. These are terrible fakes of poor craftsmanship, but they’ve had influence, just like preposterously phony pictures of mermaid skeletons or the gates of heaven.

When today’s journalists fret about the “power and peril” of AI-generated images, it often feels like what they’re really worried about is the degree these pictures will fool idiots. Smart people, after all, possess ingrained skepticism towards photos that depict unlikely or implausible things, regardless of how realistic they look. People who hold positions of trust within the mainstream media, like senior reporters and editors, will be informed enough to know that popes don’t dress in 21^st century streetwear, and even this sort of relatively basic background knowledge when paired with the simple understanding that pictures have the capacity to lie, and the professional responsibility of confirming a source, is enough to serve as a check on how much damage a false image can do within the context of a major information-spreading institution. The idea that a AI-generated photo of Trump being arrested or whatever could wind up on the cover of the New York Times strikes me as about as likely as a photo of Bat Boy ending up there. 

Those who know little, by contrast, the people who don’t know what popes wear and have a poorly-developed sense of curiosity about the images they consume, will be the ones who “fall for” deceptive AI pictures and credulously circulate them through social media. This is bad, in the sense that the spreading of ignorance and misinformation is always bad, but I do feel that the heightened quality of AI images is probably a fairly irrelevant variable in all this. 

As the popularity of the countless pre-AI fake photos on Snopes proves, sloppy production quality offers little resistance to someone inclined to believe whatever message a phony picture was created to convey. As far as nefarious technology goes, social media remains much more deserving of scorn and pessimism, since social media is the means through which the stupid are able to share their nonsense with a large community of the equally gullible.

Phony images may be easier to make with AI, but it’s always been the act of making that’s more potentially sinister than the technology used to do the making. A phony picture without ambition is pointless and easily ignored — who cares if someone creates a fake photo of the president sitting quietly in a chair that doesn’t exist? But a phony picture with ambition triggers an immediate emotional reaction — say, the president dressed as a SS officer. It is the imaginations of those who dream of dishonestly provoking their audiences into states of shock, delight, anger, or confusion, and then mobilizing them towards one cause or another that should be the priority of our concern. But worrying about the power of liars is also one of the oldest worries in human history.

A question worth asking, however, might be whether novel technology like generative AI incentivizes lying by making it fun and easy. You can go all the way back to Victorian ghost photography for examples of this sort of thing; blurred lines between the spreading of genuine misinformation and irreverent fooling around with some newfangled plaything. The moderating force becomes a creator community aware of the power they hold over those who don’t know better, and restrain themselves accordingly; the wisdom to avoid telling more lies than is absolutely necessary.